[OpenAFS] OopenAFS 1.2.13: ever increasing number of fileserver connections - h_Hold leak

Rainer Toebbicke rtb@pclella.cern.ch
Wed, 05 Jan 2005 17:51:26 +0100


Rainer Toebbicke wrote:
> Something's wrong with OpenAFS 1.2.13 as on several busy servers we see 
> an ever-increasing number of host/client connections.
> 
 >...
 >...

Digging a bit deeper it looks like the culprit is the delta 
STABLE12-callback-rxcon-protection-20031114:

GetSomeSpace_r loops over all hosts, and because of the delta lih_r 
h_Holds all of them. Later it h_Releases only the one for which it 
decides to break Callbacks.

I understand that delta's purpose was to stop the host in question from 
vanishing in h_Release_r should the HOSTDELETE flag be on.

Another possible problem with that delta: the static variable 
lih_host_held is protected by H_LOCK, but that lock is given up in the 
call to ClearHostCallbacks_r.

I'll test a fix to the problem tomorrow, I'll submit it if there is no 
smoke.


-- 
=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=
Rainer Toebbicke
European Laboratory for Particle Physics(CERN) - Geneva, Switzerland
Phone: +41 22 767 8985       Fax: +41 22 767 7155